Extrinsic calibration of mobile camera and lidar

摘要

The lidar produces two dimensional images, and multiple images are converted into a three-dimensional image (point cloud) including reflectance values as the system moves. A metric representing the difference between the two images is generated and minimized in order to determine the misalignment between the lidar and camera. After calibration the images are combined. Cubic interpolation and/or Gaussian smoothing of image pixel values may be used, as well as patch-based normalisation. The metric may be an edge based sum of squared difference function. Movement may be detected using an inertial navigation system (INS) or visual odomentry.